Types of predictive maintenance

Compare & decide · 10 min read · Updated 2026-09-01

Seven predictive-maintenance techniques, what physical fault each one actually detects, the assets they suit, the warning time they realistically give, and where each one is blind. Technique is chosen by failure mode, not by asset type.

Key takeaways at a glance

Topic	Key point
Choose the technique by failure mode, not by asset	The most common way to waste a predictive-maintenance budget is to buy a technique and then look for somewhere to point it.
The seven techniques at a glance	Technique What it detects Typical assets Warning time Vibration analysis Bearing wear, imbalance, misalignment, looseness, gear defects Pumps, motors,
Where each technique is blind	The blind spots matter more than the capabilities, because they are what a vendor demonstration never shows.
Multimodal beats single-technique	Techniques overlap deliberately.
What a technique needs before it works	Requirement Why it decides success Healthy baseline Trending is relative.
Governing standards by technique	Area Standard Condition monitoring, general guidelines ISO 17359 Data interpretation and diagnostics ISO 13379 Data processing, communication and pres

Choose the technique by failure mode, not by asset

The most common way to waste a predictive-maintenance budget is to buy a technique and then look for somewhere to point it. Each technique senses a specific physical consequence of degradation. If the failure mode you care about does not produce that consequence early enough to act on, the technique will not help however good the analytics are.

Work in this order: identify the dominant failure mode, establish whether it has a usable P-F interval, then pick the technique that sees that mode.

The seven techniques at a glance

Technique	What it detects	Typical assets	Warning time
Vibration analysis	Bearing wear, imbalance, misalignment, looseness, gear defects	Pumps, motors, fans, gearboxes, compressors	Weeks to months
Infrared thermography	Abnormal heat: loose or corroded electrical joints, overloaded circuits, refractory and insulation loss, blocked flow	Switchgear, MCCs, steam and process lines, furnaces, bearings	Days to weeks
Oil and wear-debris analysis	Lubricant degradation, water and particulate contamination, wear-metal generation	Gearboxes, hydraulics, engines, turbines	Weeks to months
Airborne and structure-borne ultrasound	Pressure and vacuum leaks, steam-trap failure, early bearing friction, electrical discharge	Compressed air, steam systems, switchgear, slow-speed bearings	Immediate to weeks
Motor current signature analysis	Rotor bar defects, eccentricity, winding faults, mechanical load anomalies	Induction motors, especially where physical access is poor	Weeks
Performance and process monitoring	Efficiency loss: fouling, wear, internal leakage, blockage	Heat exchangers, pumps, compressors, boilers	Weeks to months
Acoustic emission	Crack initiation and growth, structural defects, very early bearing damage	Pressure vessels, structures, slow-speed rotating equipment	Varies widely

Warning time is indicative and depends entirely on the failure mode, duty cycle and measurement interval — treat the column as a ranking, not a specification.

Where each technique is blind

The blind spots matter more than the capabilities, because they are what a vendor demonstration never shows.

• Vibration analysis needs stable speed and load to trend cleanly. On variable-duty machines the signal moves with the process, and without operating context you generate false alarms. It also struggles on very slow-speed shafts, where the energy is too low — that is acoustic-emission territory.
• Thermography sees surfaces, not interiors, and it is only as good as the emissivity setting and the load at the moment of the survey. A lightly-loaded circuit can look healthy and fail under full load. Reflective bare metal is the classic misreading.
• Oil analysis tells you a machine is generating wear debris, not always which component. Sampling point and procedure dominate the result; an inconsistent sampling method produces a trend of the sampling, not of the machine.
• Ultrasound is excellent at finding leaks and trap failures and poor at quantifying how bad a mechanical fault is. It answers "is there a fault" far better than "how long have I got".
• Motor current signature analysis reads the motor through its electrical supply, so driven-equipment faults arrive attenuated and mixed with supply-side distortion.
• Performance monitoring needs instrumentation you can trust. Drifting temperature or flow transmitters produce a convincing efficiency-loss trend that is purely an instrument problem.
• Acoustic emission is highly sensitive and highly noise-prone; it usually needs a specialist to separate real crack growth from process noise.

Multimodal beats single-technique

Techniques overlap deliberately. A bearing degrading on a critical pump will usually show in more than one channel: ultrasound first as friction changes, then vibration as defect frequencies appear, then wear metals in the oil, then temperature at the end when it is nearly too late. Reading two independent channels together is the cheapest way to cut false alarms, because instrument drift and process noise rarely fake the same fault twice.

This is also where analytics genuinely add value: not by inventing a new physical sensor, but by combining condition channels with process and maintenance history that a single-technique analyst never sees together.

What a technique needs before it works

Requirement	Why it decides success
Healthy baseline	Trending is relative. Monitoring fitted to an already-degraded machine encodes the fault as normal.
Repeatable measurement point	Different sensor position or sampling point equals a different signal, not a different condition.
Operating context	Speed, load and process state, or variable-duty assets produce unreadable trends.
Interval below the P-F window	Measure less often than the fault develops and the failure lands between checks.
An owner for the alarm	A route from alert to scheduled work order. Without it the programme produces reports, not avoided failures.
Certified interpretation	ISO 18436 sets qualification categories for condition-monitoring personnel per technique.

Governing standards by technique

Area	Standard
Condition monitoring, general guidelines	ISO 17359
Data interpretation and diagnostics	ISO 13379
Data processing, communication and presentation	ISO 13374
Vibration measurement and evaluation	ISO 20816 (supersedes ISO 10816)
Prognostics	ISO 13381
Personnel qualification per technique	ISO 18436
Reliability and maintenance data collection	ISO 14224
Maintenance terminology	EN 13306

Frequently asked questions

Related guides

Software that helps